Abstract
Floods are among the most destructive natural hazards, and accurate forecasting is essential for effective risk management and response. However, in many regions, sparse in-situ monitoring networks with a limited number of measurable parameters may compromise forecasting accuracy. This study focuses on the Uba River Basin in the East Kazakhstan region and investigates whether integrating ERA5-Land reanalysis data with observed hydrometeorological measurements improves hydrological forecasting, additionally providing an analysis of shared variables. Models trained solely on observed data are compared with those incorporating ERA5-Land features. Leave-year-out cross-validation across six test years reveals that while incorporating ERA5-Land generally enhances forecast accuracy, the improvement is minimal or non-existent in specific years, suggesting that dominant local hydrological fluctuations may already be well-captured by in-situ data alone. Bias analysis further indicates that ERA5-Land tends to underestimate extreme precipitation events, potentially affecting flood peak predictions. Overall, despite its limitations, ERA5-Land proves to be a valuable supplementary data source for improving flood forecasting and the findings provide insights into optimizing data selection strategies to further refine flood forecasting in data-sparse regions.